DEFINING SUITABLE HABITAT AND CONSERVATION STATUS FOR THE TUCSON SHOVEL-NOSED SNAKE (CHIONACTIS ANNULATA KLAUBERI) IN THE SONORAN DESERT Curtis M. Bradley, Center for Biological Diversity, Tucson, AZ; and Philip C. Rosen, University of Arizona, Tucson, AZ. Abstract.— The Tucson shovel-nosed snake (Chionactis annulata klauberi) is a small specialized colubrid associated with sandy loams of the elevated portions of the Lower Colorado River Valley Sonoran Desert in central Arizona. This taxon is a recently redefined subspecies based on genetic data, and may be extirpated in the Tucson region, including the type locality. A recent (USFWS 2014) decision against protecting it was based in part on an expansive definition of its geographic range and a habitat model. Here, we redefine the subspecies distribution by uniting published results of mitochondrial and nuclear DNA. We then present a new ecologically-based model of its original and current habitat using the machine learning algorithm Random Forests. The new model accurately matches known and estimated presence-absence data for this taxon, and is consistent with morphometrics and, largely, with color pattern variation. It estimates 60% less available habitat than the USFWS model. We estimate that 39% of the 1,255,946 ha (3,103,505 acres) of original habitat has been converted to urban developments, roads, agriculture, or otherwise altered non-habitat. Of the remaining 770,971 ha (1,905,108 acres), 60% is susceptible to habitat conversion in the region, with only 10.9% of habitat having current legal protection. Ongoing and projected urbanization and energy development in its flatland desert habitat present a bleak future for this subspecies. INTRODUCTION In recent decades, genetic data have greatly increased our understanding of the history and processes shaping populations, gene pools, and landscape relationships of species (Avise 2000). This revolution has reshaped concepts of speciation, species boundaries, subspecies, and appropriate targets of conservation and management (Avise and Ayala 2017). Earlier consensus regarding the “biological species concept” and subspecies have largely been supplanted by evolutionary and genealogical species concepts. Here taxa are considered species if they are on distinct evolutionary trajectories and have anciently differentiated genetically (De Quiroz 2007). Subspecies have become controversial, with some elevated to full species rank while others have been dismissed or remain in taxonomic limbo. However, adaptive subspecific differentiation may play a critical role in speciation (Gavrilets 2014), and subspecies may merit protected conservation status (Phillimore and Owens 2006). Shovel-nosed snakes (genus Chionactis) present complexities with regard to identifying species, lineages, and significant conservation entities. Four taxa are currently recognized: the relatively divergent Sonoran shovel-nosed snake (C. palarostris) and three taxa in the geographically variable C. occipitalis complex (Wood et al. 2014). These include the Mojave shovel-nosed snake (C. occipitalis) and two subspecies of the Sonoran Desert shovel-nosed snake (C. annulata): the Colorado Desert shovel-nosed snake (C. a. annulata) and the Tucson shovel-nosed snake (C. a. klauberi). The C. occipitalis complex displays complicated geographic patterns of genetic differentiation (Wood et al. 2014) making definition of taxon boundaries (Wood et al. 2014; USFWS 2014) and conservation status challenging. In the C. occipitalis complex, the head is morphologically specialized for burrowing in the sandy soils they inhabit (Ernst and Ernst 2003), more so than in the sister taxon C. palarostris. The body is slim and thus apparently adapted for rapid locomotion (Cundall 1987) compared to the stouter build of the closely related banded sand snake (Chilomeniscus cinctus), which is similarly specialized for burrowing (Ernst and Ernst 2003). Further, Chionactis in the Sonoran Desert have bold red and black crossbands over a cream or yellowish ground color, marking them as mimics of venomous coralsnakes, including the sympatric Sonoran coralsnake (Micruroides euryxanthus). In 2004, C. a. klauberi was petitioned for listing as Threatened or Endangered under the US Endangered Species Act (ESA). This petition was based on its habitat specialization in desert flats subject to agricultural conversion and urban sprawl in central Arizona, and its apparent disappearance from the Tucson region (Center for Biological Diversity 2004), which includes the type locality (Stickel 1941). The subspecies was defined based on the strong infusion of black pigment on the red crossbands, which may enhance both coralsnake mimicry (Mahrdt et al. 2001) and background matching (via flicker-fusion: Titcomb et al. 2014). Its geographic range and the presence of intergrades with C. a. annulata were described by Klauber (1951) and Cross (1979). On the basis of morphological, mitochondrial and nuclear gene analyses, Wood et al. (2008, 2014) supported continued recognition of C. a. annulata and C. a. klauberi as a subspecies with “fuzzy” boundaries due to evidence of asymmetric gene flow between the recovered genetic entities that were largely consistent with the earlier concepts of the morphologically described subspecies (Stickel 1941; Klauber 1951; Mardt et al. 2001; Stebbins 2003). However, Wood et al. (2014) did not define distributional limits for C. a. klauberi. In rejecting C. a. klauberi for ESA protection the U.S. Fish and Wildlife Service (USFWS; 2014) adopted an expansive definition of the subspecies that included all geographic regions within C. annulata with any genetic connection to C. a. annulata. Herein we define operational boundaries consistent with the complex genetic variation known in C. annulata for analysis of the conservation status of C. a. klauberi. We then model the ecological distribution and habitat extent of C. a. klauberi and re-evaluate the geographic extent of threats to its habitat and its vulnerability to extinction. MATERIALS AND METHODS Occurrence records.—Location data were coordinates for tissue sample records in Wood et al. (2014, Table S1). Delineating the geographic distribution of C. a. klauberi presented challenges associated with divergent geographic gradients in mtDNA and nuclear microsatellite data (Wood et al. 2008, 2014). We resolved this problem by using the intersection of the mitochondrial and nuclear DNA datasets from Wood et al. (2014), using individuals from microsatellite cluster C (red circles and half circles in Fig. 5a, Wood et al. 2014) that were also contained in mtDNA clade C. This resulted in 44 locality points from which we added verified museum and published locality records that were within the previously delineated range of C. a. klauberi (Mardt et al. 2001) but not sampled by Wood et al. (2014). This included records from Santa Cruz Flats, Pinal County, and Avra Valley near Tucson, Pima County (including the type locality) that could be mapped with a precision of 0.1 miles or better. We placed a grid of 1 km square cells over the points and randomly selected one occurrence per cell to remove repetitive records that likely reflect sampling bias and to reduce autocorrelation. The resulting dataset represented 53 C.a. klauberi “presence” records (Fig. 1b) that were used as model inputs. Absence records.—We constructed model inputs of “absence” records based on paved road segments (the only way this subspecies was effectively sampled) that have been intensively searched without yielding records of C. a. klauberi. Points were placed along these road transects at 10 km intervals, excluding all areas within 10 km of a record of a C. a. klauberi record. Road transects are often in areas of low slope gradient so we placed additional absence records on popular mountain ranges near Phoenix where there are no records of Chionactis. We placed an additional eleven absence records in western Arizona at localities representing microsatellite clusters A or B (Wood et al. 2014). The absence dataset thus included 74 locality records (Fig. 1) yielding a balanced presence/absence dataset for use in our model. Environmental Predictors.—We assembled 38 explanatory (x) variables that literature (Klauber 1951, Cross 1979, Stebbins 2003, Ernst and Ernst 2003) and our field experience suggested as potentially important to the species (Table 1), with particular attention to soils. We extracted percent sand, silt, clay, rock fragments, and available water storage from the USDA’s STATSGO2 database for Arizona (STATSGO 2016). We did not use the finer scale SSURGO database because it has significant data gaps and discontinuities within our study area. Soil map units were converted into a 90 m resolution grid for use in our model that appropriately represented the resolution of the original vector data. All subsequent raster datasets mentioned herein were resampled to match this 90 m resolution using bilinear interpolation for continuous data or majority filter for discrete data. We represented vegetation with the Biophysical Settings, Vegetation Cover, and Existing Vegetation Type (EVT) datasets from LANDFIRE (LANDFIRE 2014). The Biophysical Settings maps depict the vegetation likely present before Euro-American settlement and was used in our model because many observation records pre-date valley conversion to non-habitat. The Vegetation Cover dataset was modified to indicate the percent shrub cover. The EVT maps were used to classify areas that have been converted to non-habitat